perf: cut per-frame cost in the websocket receive pipeline (release 0.1.4)

[joeblau]

Bumps to 0.1.4. Removes duplicated machinery on the inbound frame path and the per-frame async wrapper on the heaviest channel, then restores two behaviours that regressed against 0.1.3.

Receive pipeline

change	win
Socket no longer re-boxes every frame into a MessageEvent for the SDK's own consumer	1.42×, 305 ns/frame
fastAssetCtxs decodes via native inflateRawSync and delivers in the dispatch tick	7.6× delta, 1.8× snapshot
Lone listener stored unboxed; copy-on-write array on promotion (no snapshot copy/clear)	2.6× at 8 listeners, 1.18× at 1
Dispatcher reuses the id it already built by concatenation as the wire frame	~54 ns/subscribe

ReconnectingWebSocket was allocating a MessageEvent and paying a full EventTarget.dispatchEvent per frame purely to hand it to HyperliquidEventTarget — the exact per-frame allocation cost this SDK removed from _emit, still sitting one layer upstream. The public message event is still dispatched whenever anything outside the package listens.

Measured against the committed baseline: webData3_frame_dispatch_e2e −38%, l2book_dispatch_50_coins −21%, reconnect_resubscribe_burst −16%, subscribe_200_coins −16%, spotState_frame_dispatch_e2e −12%, ws_request_round_trip −5%, clearinghouseState_frame_dispatch_e2e −3%. Large frames move least because JSON.parse dominates them and the SDK can't touch that.

DecompressionStream remains the browser/RN fallback; both decode paths were differentially tested and agree on all 14 cases including 8000-coin multi-chunk payloads and every invalid input.

Fixes

• Wrong-frame delivery on fastAssetCtxs. e.detail was read from the recycled event shell inside a queued continuation. Once frames arrived faster than inflate completed, the continuation read a later frame's payload — deltas silently dropped, others delivered twice, corrupting downstream price state. Reproduced across three arrival patterns before fixing.
• Frame delivered after unsubscribe. A listener unsubscribed by an earlier listener mid-dispatch still received the in-flight frame. 0.1.3 matched EventTarget here; the snapshot-buffer rewrite broke it. A generation counter restores it at one integer compare per listener.
• Unvalidated vault address. createL1ActionHash is publicly exported and took vaultAddress with no runtime validation — a 0x${string} template type constrains neither charset nor length at runtime, and not at all for JS callers. A malformed address silently hashed to different bytes than the caller described, and a 32-byte address hashed identically to its 20-byte truncation. Validation restored as it was at 0.1.3.

Tests

1652 pass / 0 fail. New coverage for the listener registry, the frame hook and public message behaviour, arrival order when synchronous and queued decodes interleave, plus two perf scenarios for the fastAssetCtxs decode path, which previously had none.

Reviewer notes

• The Performance job is expected to flag this. It fails closed when the perf suite's source fingerprint differs between base and head, and this PR adds a scenario file. That's the designed "flag for explicit review" path, not a regression.
• baseline.json still needs re-recording on a quiet machine. The committed one was recorded with dirty: true and its across-the-board gains are largely session drift — scenarios over untouched code (http_request −24%, parse_l2_book −15%) moved as much as the real wins. Its rme also inflated sharply on several scenarios, which widens the local gate's tolerance. It's now also stale against the new suite fingerprint. I did not re-record it: this machine was at load average 10 during the run, which would bake in the opposite bias.

🤖 Generated with Claude Code